• Artificial intelligence (AI): computer systems that perform tasks that usually require human judgement or decision-making.  
• Machine learning (ML): a way of building AI systems by training them to find patterns in data, rather than programming them with fixed rules. 
• Large language models (LLMs): a type of ML model trained on large amounts of text to learn statistical patterns in language, including how words and word patterns commonly relate to one another. LLMs convert words or parts of words into numbers so they can process text. They can be used to interpret natural language ‘prompts’ and predict what text is likely to come next, producing fluent and compelling responses that may not be accurate.    
• Generative AI (GenAI): a type of AI tool that uses LLMs and other powerful ML models to generate new content – such as text, images or audio. This is the main type of AI covered in this guidance. For simplicity, this guidance uses the term ‘AI’ when referring to GenAI, unless otherwise stated. 
• Agentic AI: AI systems that can make decisions and take actions to complete a task with limited human supervision. Unlike traditional AI tools that simply respond to instructions, agentic AI can carry out multiple steps on a user's behalf. For example, it might review a client enquiry, draft a response, schedule a follow-up meeting and send notifications without needing a separate instruction at each step. 
• Prompts: the natural language instructions used to instruct many GenAI tools. Clear, precise and well-structured prompts will usually produce better quality outputs. 
• AI hallucinations: GenAI uses statistical models to predict outputs – it does not understand the meaning of words or images, and it cannot tell what is true or false. That is why GenAI can produce errors or ‘hallucinations’ and why lawyers must always verify its outputs. Hallucinations may include inaccurate or fictitious citations and quotations, or citations and quoatitions that do not support the key proposition. Hallucinations may occur even with GenAI tools designed for legal practice and tools using Australian legal content or a firm’s own content.  
• Retrieval augmented generation (RAG): a way of configuring a GenAI tool so it searches specified source documents or databases at the time it is asked a question and uses that information to generate an answer. This can help reduce errors, show where information has come from, and make outputs easier to check. However, it does not remove the risk of hallucinations, omissions or misinterpretation.  

Many AI tools work online. These tools send your prompt and any uploaded files to a remote server (generally controlled by the AI provider) before the response is sent back to you. When this information leaves your device there is a security risk unless the provider gives security guarantees. Client information, and other confidential, personal or private information must not be used with tools that don’t provide appropriate security guarantees. 

A secure AI tool will provide clear information and guarantees about privacy, cybersecurity and how your information will be used, including whether prompts or uploaded material can be accessed by the provider or used to train AI models. We set out the security guarantees to look for in our AI checklist for law practice leaders and principals.  

AI tools can be standalone products or built into software your practice already uses. Do not assume a tool is secure because it is paid, embedded in existing software, or described as an enterprise product. Always check the specific security guarantees and contractual protections for the tool. 

Some AI tools run on your device, or on servers controlled by your organisation. This is often referred to as ‘on-device’, ‘on-premises’ or ‘self-hosted’ processing. The security of these tools depends on the overall security of your devices, and your organisation’s servers and network.  

Read more about our expectations for law practices when securing work devices

Scenario 

You want to create or improve a generic draft email that does not include any client-specific information. The email explains the next steps in a common type of matter, such as the steps needed to complete settlement after a contract for sale has been signed.  

How you could use AI 

You prompt a free AI tool to generate a first draft, making sure you don’t include any client information, confidential information or matter-specific details in the prompt. 

You review the AI-generated text, make necessary adjustments to the tone, structure and wording, and check that it accurately reflects the law and legal processes. You then add any client-specific information outside the AI tool and review the final email carefully before sending it. 

As you commonly send this type of email, you save the generic version without any client information as a template for future use. 

Tips 

• Never enter client information, or other confidential, private or personal information into an AI tool unless it is a secure tool that has been approved for use with that information by your law practice. 
• Proofread and verify all AI-generated outputs – even for simple tasks like routine emails – to ensure the content is accurate, complete and uses appropriate language. 
• AI can help make routine emails and communications clearer (e.g. reduce legal jargon), better structured and quicker to draft.  
• If you have access to a secure AI tool you may be able to use it to draft a client email from scratch using relevant matter or client information, or to review a draft email that contains client or confidential information to improve clarity, structure, tone or readability. Still check the final email carefully before sending it. Make sure it is accurate and reflects your professional judgement. 

Scenario 

You need to create a standard form template letter for onboarding new clients for a particular type of matter. It needs to explain, in general terms, the information and documents they must provide, their obligations as a client, and the next steps.  The template needs spaces to fill in details about the client, the matter, and the scope of work.

How you could use AI 

You use a free AI tool to draft a template letter by creating a prompt, describing the context of the letter and the structure and tone you want.  

You review the output, making sure it is consistent with your practice’s actual processes, then provide further prompts to refine the tone, structure and content. 

You review the final AI-generated content carefully and make any necessary edits, then put the text into your firm’s letterhead. 

Tips 

• Always review and verify AI-generated output before using it. Make sure it doesn’t contain errors or inappropriate language. 
• You don’t need to use a secure AI tool to generate a generic template provided you do not enter any client information or other confidential, personal or private information. 
• If the template needs to include commercially sensitive, or confidential, information, either use a secure AI tool or add that content manually at the end. 

Scenario 

Your client has received a lengthy letter of demand and has retained you to advise on and respond to the letter. You need to prepare a reply that addresses each issue raised in the letter, sets out several points of law, and explains how they relate to your client’s particular circumstances. 

How you could use AI 

To save time producing a final draft, you decide to use a secure AI tool to help edit and refine your work. 

You prepare a rough first draft of the letter, setting out the points you want to make. You draft a prompt for the GenAI tool, attaching your draft, the original letter of demand, and setting out your client’s relevant circumstances. You instruct it to provide suggestions on how to improve the letter and identify any gaps.   

You carefully review and incorporate the tool's suggestions. You review the final product, making sure it has not introduced any errors into the document, confirming each statement of law and fact in the letter is correct, and verifying every reference to a legal authority, including any citations, by reading the authority in full. 

You send the letter to your client for review and make a file note of how you used AI to help prepare it and how you verified the content generated by AI.  

Tips 

• Do not depend on AI to determine your legal strategy or assess the merits of your client’s position – these are matters for your professional judgement. 
• If the AI tool you have access to is not secure, you must not include any client or other confidential, personal or private information in your prompt. This may mean you can only use the tool for parts of your document (e.g. drafting a paragraph that sets out a principle of law in general terms). 
• Take extra care with information subject to additional protections – such as a Harman undertaking – or a suppression order. Make sure the security guarantees provided by any AI tool used are sufficient to meet those additional obligations. 

Scenario 

You will be representing a client in court in an area of law that is familiar to you, but you have not prepared submissions for this type of hearing before. 

How you could use AI 

You use a free AI tool to brainstorm an approach, identify relevant legislation and case law, and develop a structure for your submission. You write a prompt that asks the tool to identify possible counterarguments, gaps in your reasoning, or alternative ways to frame the argument.

As the tool is not secure you only enter anonymised, hypothetical scenarios that don’t identify your client or include any client or confidential information.

You take the tool’s suggestions as a starting point and then do your own legal research, verify any points of law it suggested, consider whether its counterarguments or alternative framings are useful, and draft the submissions yourself. 

Tips 

• Always identify and comply with any relevant court practice notes, rules or guidance on the use of AI in court proceedings. These may include requirements to disclose the use of AI or restrictions on using it to prepare affidavits and other evidence. 
• When working on litigation-related materials, even with a secure tool, be mindful of Harman undertakings and any other restrictions on how information can be used. 
• A law-specific tool that draws on trusted sources – such as legislation, case law databases, or your practice’s own precedents and documents – may produce stronger and more traceable outputs than a free, general-purpose tool.  
• Whichever tool you use, you must verify the output and take particular care with citations, statements of law and references to legal authorities — AI can produce false citations, inaccurate quotations and other serious errors.  
• You are responsible for exercising your own forensic judgement when advising clients and preparing submissions. Do not rely on AI as a substitute for your own assessment and analysis of the client’s needs, circumstances, case strategy, or the authorities to put before the court. 

Scenario 

You have prepared a detailed memorandum of advice for your client. You think it could be shorter and written in language that is easier for a non-lawyer to understand. Revising it manually would take a significant amount of additional time and could increase costs for your client.

How you could use AI 

Your law practice has implemented a secure AI tool that integrates with your word processing software and is approved for use with client and confidential information.  

You prompt the tool to shorten the document and simplify the language using plain English. You review the AI-revised document carefully to ensure it is accurate and that the tool has not introduced any errors, omissions or changes in meaning before you send it to your client. 

Tips

• If you don’t have access to a secure AI tool, you could extract parts of the document that don’t include client information and use a free tool to refine them – but you must fully de-identify any text or document (including metadata). 
• Review all AI-generated outputs closely, even if the tool has only edited, refined or formatted a document. AI can still ‘hallucinate’, introduce errors, omit important nuance, change the meaning or tone of your advice, or introduce inappropriate assumptions or biased language.   

Scenario 

You are meeting with a client to discuss their matter and potential legal options. You want to focus on the discussion while keeping an accurate record, without increasing your client’s costs by having a paralegal attend to take notes. 

How you could use AI 

Your law practice has implemented secure meeting software and approved it for use with confidential information. The software has built-in recording and AI transcription features. 

You explain to your client that you have access to a secure AI tool to record and transcribe the meeting, including how the recording and transcript will be stored and handled, and ask for their permission to use the tool. With your client’s consent, you record and transcribe the meeting. The software also produces a AI summary of the transcript. 

Shortly after the meeting (while it’s still fresh in your mind) you review the AI-generated transcript and summary and correct any errors or omissions.

You save the verified documents to your client’s file and make sure the meeting recording is stored securely and deleted when it’s no longer needed. 

Tips   

• You should obtain your client’s consent before using AI to record and transcribe meetings. 
• Only keep recordings if needed and store them securely to avoid breaching confidentiality or privacy obligations – meeting software may save recordings in locations with inappropriate default access or retention settings.  
• Keep in mind that AI transcription can be less accurate when audio quality is poor, multiple people speak at once, technical terms or names are used, or when speech is in a language other than English, spoken with an accent or in a less common dialect. 
• Note the key points raised during the meeting so that you can check the AI transcript and summary against your notes. Review the transcript and summary shortly after the meeting to make it easier to identify and correct errors and omissions. 

Scenario 

You have a client matter involving documents written in another language that you cannot read fluently. You want to understand the documents’ general contents so you can decide whether you need certified translations to support your advice or use as evidence. 

How you could use AI 

Your law practice has access to a secure AI tool that has translation features and is approved for use with client and confidential information. 

You upload the documents to the tool and prompt it to generate initial English translations. You review the AI-translated text carefully and consider whether you need the content verified by a certified translator or advice from a lawyer fluent in that language.  

When you save the AI-translated documents to the client file, you clearly label them as AI-generated translations of the original documents. 

Tips 

• AI translation can help you understand a document’s general content. However, if you need to rely on a translation (e.g. as evidence, for legal advice or formal communications), it should be sent to a NAATI-certified translator for a professional translation. 
• If you use client documents with an AI tool that is not secure, you risk breaching your duty of confidentiality. Only use publicly available information (e.g. websites or public documents) when using free online translation tools. 

Scenario 

You are working on a matter where your client has provided you with a large collection of potentially relevant legal documents and evidence. You have advised your client that certain documents would assist with the matter, and they believe those documents are somewhere in the collection, but manually reviewing every document to find the relevant ones would take significant time. 

How you could use AI 

Your law practice has implemented a secure AI tool that integrates with your document or file management software and has been approved for use with client and confidential information. 

You propose using this tool to your client, carefully explaining the risks associated with using an AI tool instead of reviewing each document manually. You also explain the difference in cost to your client. Your client agrees to you using the tool to search for the documents. 

You prompt the tool to search the document collection using natural language descriptions of what you are looking for. You then review the documents and passages it identified as potentially relevant to confirm they are relevant, and to check whether any summaries it generates are complete and accurate for your purposes.  

For large document collections (for example in discovery for litigation matters) consider whether a purpose-built e-discovery or technology assisted review tool is more appropriate than a general AI tool. These tools are designed to search, review and manage large volumes of documents, and may be more effective.  

Before using an AI tool this way, make sure it has access to the documents you need to search. 

Tips 

• An AI tool can miss things and may not find every relevant document. You should always take steps to verify independently that you have located all the documents you need, including by reviewing key source documents yourself, and/or consulting further with your client. 
• AI tools used to search documents can still ‘hallucinate’ or make errors about their content (such as missing important context or identifying incorrect dates, deadlines and names). Always confirm the information against the source documents. 
• Make sure your client is properly informed about the risk and consequences of missing relevant information before they agree to use an automated process. 
• Your client may instead want to search for the documents themselves – make sure they also understand the risks if they attempt to do this with their own (potentially less secure or less capable) AI tool. 

Scenario 

A law reform commission has released a long report which may be relevant to your future work. You want to understand its key findings and recommendations and then send a short summary to interested colleagues. 

How you could use AI 

You use a free AI tool to generate a summary of the report. You prompt the tool to focus on the parts most relevant to your work (e.g. by providing general context about the nature of your work or asking it to focus on information relevant to a specific area), and to provide page references for the material it identifies. 

You review each finding summarised by the tool and check the cited pages in the report to confirm the summary is accurate. 

You then prompt the AI tool to draft an email setting out the key findings and recommendations. You review the draft email, make any necessary changes, and send it to your colleagues. In your email, you disclose that AI was used to prepare the summary, note its limitations and explain how you verified it.

Tips 

• Because the report is public information, you don’t need a secure AI tool to summarise it. 
• Uploading copyrighted materials such as journal articles or books may infringe copyright, so check before doing so. 
• If you share an AI-generated summary with colleagues, you should disclose how it was produced. Do not present it as definitive, unless you have fully reviewed the source material and verified the summary yourself.  
• For background reading that you will not rely on in your work, it may be enough to review only the relevant cited sections of the source document – provided you make clear the extent to which you have verified the summary.  
• For client matters, only rely on the summary after reading the source document in full and verifying its accuracy. 
• AI tools generally have limits on the length of text that can be used with them in a single prompt or conversation, or over a period of time. These limits may depend on whether (and how much) you are paying to use the tool. Attempting to summarise very large documents could exceed these limits.  

Scenario 

You take over a matter shortly before a scheduled pre-trial mediation. The file contains a large volume of documentary evidence and facts in dispute. To prepare for the mediation, you need a summary of the legal issues, a chronology of events, and an index of the documentary evidence.

How you could use AI 

You first review the client file and the source documents yourself. You then use a secure AI tool integrated with your case management software to help prepare the additional materials you need.  

You prompt it to review the entire client file and: 

• summarise the legal issues, the case strategy, and the other party's position 
• generate a chronology of events linked to the relevant evidence 
• generate an index and summary of the relevant documents in the file. 

You carefully review the materials the tool produces. You make sure all the relevant issues are identified and described accurately, and that the chronology and document index are complete and correct by cross-checking them against the source materials. You also clearly label the materials as being AI-generated. 

Once you have reviewed and revised the materials, you share them with the colleagues attending the mediation – and explain that AI was used to help prepare them, including how you verified the outputs.

Tips   

• Don’t assume AI-generated summaries, chronologies and analyses are accurate – the risk of ‘hallucinated’ errors or important information being omitted means these materials must always be verified. 
• If colleagues rely on materials you prepared using AI, make sure they understand how they were generated and how you verified them. 
• AI tools will often have a limit on the number of documents or attachments that can be used in a single prompt or conversation, as well as limits on the total amount of material that can be used with them. Using very large inputs can also reduce the accuracy of generated outputs. If you’re working with large collections of files, make sure you’re using an appropriate tool and understand its limitations. 

Scenario 

You are advising a client on an employment dispute. You need to review their payslips to confirm whether they have been paid correctly over several years. They provide the payslips to you as scanned images in PDF files.

How you could use AI 

Your law practice has implemented a secure general-purpose AI tool approved for use with confidential information. The tool can extract text and tables from scanned documents. 

You upload the PDF files and prompt the tool to extract the key information from each payslip into a spreadsheet, with one row for each payslip. 

You then compare the spreadsheet against the original payslips to confirm that the extracted figures and other information are accurate. 

Tips  

• If you do not have access to a secure AI tool, you may still have other secure software (such as a PDF editor) that can convert images into selectable text or documents using OCR (optical character recognition). 
• Always verify information extracted with AI. AI tools can make mistakes, omit important details, or reformat, infer or summarise information (instead of just extracting it), which can introduce errors. 
• To improve accuracy, your prompt should specify the exact fields to extract and the required spreadsheet format.  

Scenario 

You have a client with a legal problem in an area of law you are familiar with, but the matter raises a more obscure issue you have not researched in depth before. You need to research the issue and identify relevant authorities before you can advise the client. 

How you could use AI 

You decide to use an AI tool designed for legal research. Because it does not provide adequate security guarantees, you do not enter any client information. 

You ask the tool general questions about the legal issue and prompt it to identify relevant authorities. Where needed, you use only anonymised or hypothetical information. 

You use the tool’s output as a starting point only. You check every citation and read the relevant source material before relying on it. You also consider whether to carry out further research without an AI tool to find any potentially missing, more current or more relevant authorities. 

Tips 

• AI can support legal research, but it is not a substitute for reading authoritative sources and applying your own professional judgement. Always check the source material in full.  
• Always verify AI-assisted research yourself. Important information and nuance may be missing or inaccurate That risk also applies to specialised legal AI tools. 
• Do not rely on another AI tool to verify the AI-generated content – you must do this yourself. 
• If a tool links its answer to legislation, cases, commentary or your practice’s own materials, use those links to check the answer against the source. Do not assume source-linked answers are correct. 
• You can use a free, general-purpose AI tool for legal research. However, tools designed for legal research will generally produce stronger and more traceable results because they may draw on legal databases.  
• If a legal research tool does not provide appropriate security guarantees, do not enter client information or other confidential, personal or private information. Use general, anonymised or hypothetical facts only.